RSS-Based Localization using Deep Learning Models with Optimizer in LoRaWAN-IoT Networks

As a key Internet of Things (IoT) enabler, Long Range Wide Area Networks (LoRaWAN) have attracted a lot of attention in industries as well as research. Improved localization techniques are needed for large-scale applications in LoRaWAN-based IoT networks. When determining the precision of estimated location devices by signals from transmitters of reference sites, RSSI (Received Signal Strength Indicator) is crucial. The position of the devices or objects can be determined by comparing the RSS (Received Signal Strength) readings from the known reference sites. In this paper, a DL (Deep Learning) model such as Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), Multilayer Perceptron (MLP), Radial Basis Functions (RBF), Self-Organizing Maps (SOMs) and Generative Adversarial Network (GAN) with the Adam optimizer is used to design a RSS-based indoor localization. All these DL models are evaluated and compared in terms of accuracy, error, R-squared, and evaluation time. The simulation results shows that the proposed RSS-based localization along with DL techniques and Adam optimizer has shown better results than the traditional DL models. The experimental results also analyzed the comparison of variations in a number of epochs and mini-batch sizes.